LETTERS TO NATURE

Episodic multiregional cortical coherence at multiple frequencies during visual task performance

Steven L. ~ressler*, Richard Coppolat & Richard ~akamural

* Center for Complex Systems, Florida Atlantic University, Boca Raton, Florida 33431, USA t NIMH Neuroscience Center, St Elizabeths, Washington DC 20032, USA 2 Laboratory of Neuropsychology, National Institute of Mental Health, Bethesda, Maryland 20892, USA

THF: way in which the brain integrates fragmentary neural events at multiple locations to produce unified perceptual experience and behaviour is called the binding problem'.z. Binding has been pro- posed to involve correlated activity at different cortical sites during perceptuomotor behaviour*" particularly by synchronization of narrow-hand oscillations in the y-frequency range (30-80 Hz)".7. In the rabbit olfactory system, inhalation induces increased y-cor- relation between sites in olfactory bulb and cortexx. In the cat visual system, coherent visual stimuli increase y-correlation between sites in both the same and different visual cortical areas' 12. In monkeys, some groups have found that y-oscillations transiently synchronize within striate cortext3, superior temporal s ~ l c u s ' ~ and somatosensorimotor corte~'~.'". Others have reported that visual stimuli produce increased broad-band power, but not y-oscillations, in several visual cortical area^'^.'". But the absence of narrow-band oscillations in itself does not disprove inter- regional synchronization, which may be a broad-band phenomenon. We now describe episodes of increased broad-band coherence among local field potentials from sensory, motor and higher-order cortical sites of macaque monkeys performing a visual discrimina- tion task. Widely distributed sites become coherent without involv- ing other intervening sites. Spatially selective multiregional cortical binding, in the form of broad-band synchronization, may thus play a role in primate perceptnomotor behavionr.

We recordcd local field potentials (LFPs) from up to 15 cortical sites in one hemisphere of adult rhcsus macaque monk- eys (Fig. 1) performing a visual pattcrn discrimination task and found task-related multircgional synchronization over the entire frcqucncy rangc cxamincd. Our rcsults are compatible with thc hypothcsis that pcrccptuomotor intcgration involves the broad- band binding of widespread cortical sites.

A total of 5,827 corrcctly pcrformcd trials from three monkeys (G.E., T.1. and L.U.) were studied, including 2,974 trials with a rcsponsc (GO) to one visual pattcrn typc and 2,853 trials with the response withheld (NO-GO) to another type. The trans- cortical field potential a t each site was localized by diffcrcntial recording with a surface-to-depth bipolar clcctrode. The amplifier reduced common contributions to the two electrode tips by morc than 10,000 times, eliminating propagated fields from sourccs morc than a few millimctrcs away. The precision of localization was demonstrated in LFP averages which, follow- ing the v~sual stimulus, showcd visual receptive field specificity in striate and prestriatc regions but not in others. All LFP avcr- ages had a high degree of day-to-day reliability ovcr multiple recording sessions covering weeks to months.

Cohercncc and phase spectra, the frcqucncy domain equiva- lent of temporal cross-correlation, were used to test for intcr- site synchronization over a range of frcqucncics'y. Multiple cpisodes of elevated cohcrcncc, lasting 50-200 ms and involving multiplc frequencies, were observed after the stimulus (Fig. 2). Changcs in cohcrcncc ovcr time generally followed the same pattcrn at multiple frcqucncics. The rangc of peak times for these

NATURE . VOL 3 6 6 . 11 NOVEMBER 1 9 9 3

Ant

Ant.

Post.

Post.

FIG. 1 Locations of recording sites in monkeys G.E., T.I. and L.U. At each site, the signal from a transcortical bipolar electrode (51-pm-diameter Teflon-coated platinum wires: 2.5 mm tip separation, 0.5 mm exposed surface per tip, less-advanced tip at the pial surface) was differentially amplified by a Grass model P511J amplifier ( -6 dB at 1 and 100 Hz, 6 dB per octave falloff), digitized at 200 samples s ', and stored in 12- bit digital form. METHODS. Experiments were done at the Laboratory of Neuropsychol- ogy at NIMH. Animal care was in accordance with institutional guide- lines. Surgical procedures were as described in ref. 26. Monkeys were trained to accept chair restraint readily, to respond (GO condition) to one visual pattern type (line or diamond), and to withhold responding (NO-GO condition) to the other. Equiprobable GO and NO-GO trials were randomly presented in 1,000-trial sessions lasting about 45 min (inter- trial interval randomly varied between 0.5 and 1.25 s). The monkey initiated each trial by depressing a lever with the preferred hand. The head was held 57 cm from the display screen, with 1 deg cm-' visual angle. The computer-generated visual stimulus was presented for 100 ms through a silenced piezoelectric shutter, beginning about 115 ms after data collection began. On GO trials, a water reward was provided 500 ms after stimulus onset if the hand was lifted w~thin 500 ms. On NO-GO trials, the lever was depressed for 500 ms. The stimulus set consisted of four diagonal patterns wlth two dots (9 mm per s~de) at opposite corners of an outer square, 6 cm per side, and two dots at opposite corners of a concentric inner square, 2 cm per side. The outer and inner dots were slanted in the same direction in two patterns (lines), and in opposite directions in the other two (diamonds). Thus, no single dot could be used to discriminate between lines and diamonds.

LETTERS TO NATURE

FIG. 2 Coherence time-series for two different site pairs in monkey G.E. Time-series for 7 frequency bins (12.5, 25, 37.5, 50, 62.5, 75 and 87.5 Hz) are dis- played in each plot. Left, GO condition (488 tr~als). Right, NO-GO condition (482 trials). a, There is a large increase in broad- band coherence between striate

Striate - motor 1 , I

episodes extended from 90 ms after stimulus onset to beyond response onset.

Coherence was significantly above baseline in at least one fre- quency bin during the analysis period for about 66% of the site pairs in G.E., 93% in T.I. and 90'%1 in L.U. Data became signifi- cant in all bins by about 37% of the site pairs in G.E., 56'1/;, in T.I. and 20'%, in L.U. The sites involved were in striate, prestriate, inferotemporal, superotemporal, posterior parietal, somato- sensory, motor and frontal cortices. In no case were two sites coherent with zero phase variance over trials, indicating that

and motor sites at the time of the Striate - parietal response in the GO h I

synchronization was not simply an externally imposed artefact. Of the site pairs having significant coherence in all bins, 50%

had a significant difference between coherence in the GO and NO-GO conditions. The mean time of significant between-condi- tion difference (285f 24 ms) occurred shortly after the mean rcsponsc onset time (279 f 15 ms). Thus, between-condition diffcrences in coherencc (Fig. 20) appeared to be largely related to response processing. Signal power (Fig. 3) also significantly differed between conditions at 60i%, of the sites (mean time of significant difference = 319 f 20 ms). Power was significantly different for a t least one site in 96% of those site pairs having significant between-condition coherence differences, and for both sites in 54'%1 of the pairs. Like coherence, changes in power were generally similar across frequencies, with no evidence of a concomitant low-frequency decrease and y-frequency increase as has been previously ~ u g g e s t e d " ~ " ~ ~ ' .

condition. b, The same striate site shows elevated 0.8

broad-band coher- ence with a parie- a~

tal site between 100 and 200 ms post-stimulus. (The 2 o., striate slte does u not show signifi- cant coherence 0.2 with other inter- vening sites.) METHODS. Span- o

Striate - motor

Go -

-

, 1 -

I ', - ' ,

I/ I

--

Striate - parietal

nlng the period - 6 m u , 100 200 300 400 from 115 ms be- Onset Time (ms) fore until 500 ms after st~mulus onset, coherence and phase spectra were computed at each of a series of 80-ms time windows. The spectra had frequency bins at 12.5 Hz ( 1 cycle per 0.08 s) intervals. Coherence time-series, formed for each frequency bin as the time window was stepped, point by point, across the analysis period, were computed for all pairwise combinations of sites for both GO and NO-GO conditions in each ses- sion. All GO or NO-GO trials from a session were used to estimate each site pair's coherence. For statistical evaluation, the Fisher z-transform was first applied to the bounded coherence distribution to yield a roughly normal one. Six sessions were examined: two from each of the

three monkeys. Coherence values were considered significant when exceeding the 95% confidence limit for the mean coherence over all site pairs from both conditions in the analysis period of a session. Two- way analysis of variance was done for each site pair (response condition and time as major effects, frequency bins as cases). Significance of between-condition difference was determined at the 95% level after correction for multiple comparisons by the Bonferroni method. Scheffe comparisons were later made to determine the time points showing significant between-condition differences.

Single trials were identified in which the L F P waveforms from two sites were synchronized during the time when their coher- cnce was clcvated (Fig. 4). Like coherence ovcrall, cross-correla- tion on these trials was also elevated because of waveform synchronization (Fig. 4u). Synchronization of y-frequency com- ponents was not apparent bccause of their relatively small ampli- tude. But when y-frequency waveforms werc isolated by digital filtering, thcir synchronization was also evident (Fig. 4h).

Elevated cohercnce, rather than simply appearing bccause of common activation of multiple cortical areas by an imposed stimulus, manifested spatially and temporally complex patterns in the period after stimulus presentation, when the monkeys had to discriminate successfully between two visual forms and then perform correctly. These task-related patterns of multiregional synchronization arc consistent with those previously observed below 10 Hz in human scalp-recorded averaged event-rclated potentials" 24. But thc finding of broad-band elevated cohercnce during the motor response differs from those of studies in monkey"." and catz5 reporting narrow-band y-synchronization in relation to motor activity.

The rapid rise in coherence over a broad frequency range during specific stages of task processing may rcflect instanta- neous inter-regional cortical coupling on multiple timescales, providing a flexibility to cortical information processing not available through operation on a single characteristic timescale. The present results suggest that synchronization can potentially

NATURE. V O L 3 6 6 . 11 NOVEMBER 1993

LETTERS TO NATURE

FIG. 3 Average power time-series for the striate and motor sites of Fig. 2a. Time-series for 7 frequency bins (12.5, 25, 37.5, 50, 62.5, 75 and 87.5 Hz) are dis- played in each plot. Power was down nearly to the size of a single amplitude step at 100 Hz. The striate site (a), but not the motor site (b), shows elevated power after stimulus onset. Both the striate and motor sites show eleva- ted power after response onset in the GO condition (left), but not at this time in the NO-GO condition (right). Averages were over 488 GO and 482 NO-GO trials. METHODS. Span- ning the period from 115 ms be- fore until 500 ms after stimulus onset, power spec-

Striate

M e a n Response Onset I

J - ~ ~ ~ ~ ~ , ~ ~ 100 2Q ' 300 4Q & Onset

Motor

11 -

Response Onset

! i ~ ~ ~ ~ & ~ ~ ~ ~ 100 200 ' 3 0 0 400 500 Onset

Time (ms)

Striate

5 s t u u s 100 2k1 3b 400 d o Onset

Motor

~ - ~ o ~ ~ ~ , , + ~ ~ ~ ~ 100 zio 300 4 i0 sbo Onset

Time (ms) tra were computed by the Fast Fourier Transform for the same series of 80-ms time win- all sites for both GO and NO-GO conditions and averaged over the trials dows used in computing coherence. The power spectra also had fre- of each session. Similar statistical procedures were used in determining quency bins at 12.5 Hz intervals. Power time-series were computed for significance as for coherence.

a Unfiltered

3 f i

Time (ms)

FIG. 4 a, Unfiltered waveforms from the striate and motor sites of Fig. 2a for a single GO trial, with their squared cross-correlation plotted below as a function of time. The period of elevated cross-correlation in this trial corresponds to the broad-band peak in coherence (over all GO trials) in Fig. 2a. b, The waveforms in a after digital filtering to show activity in the y-frequency range. The period of synchronization (marked by the bar), corresponding to the cross-correlation peak below, demon- strates that the similarity in single-trial temporal structure extends to the y-range. The maximum squared cross-correlation value (0.86 at

Time (ms)

295 ms) exceeded the upper 95% confidence limit (0.50) for the mean value of the time-series. METHODS. Cross-correlation functions were computed on single-trial records for the same 80-ms time windows used to compute coherence. Cross-correlation time-series were then constructed using the maximum of the squared cross-correlation function in each window. The y- frequency signals were obtained by time-domain convolution with a bandpass (-6 dB at 30 and 8 0 Hz, 864 dB per octave falloff) digital filter.

NATURE . VOL 366 . 11 NOVEMBER 1993

LETTERS TO NATURE

occur between any cortical areas, and provide direct evidence in support of theories',' proposing that multiregional bind- ing underlies whole-brain functions such as perception and action.

Received 4 May; accepted 9 September 1993.

1. Damaslo, A. R. Neural Comput. I, 123-132 (1989). 2. Slnger. W. A. Rev. Physiol. 55, 349-374 (1993). 3. von der Malsburg, C. & Schne~der, W. Biol. Cybern. 54, 29-10 (1986). 4. Abeles. M. Local Cortrcal Crrcuits (Springer. Berlln, 1982). 5. Engel. A. K et a1 Trends Neurosc~. 15, 218-226 (1992). 6. Bressler, S. L. Trends Neuroscr. 13, 161-162 (1990). 7. Crlck. F. & Koch, C. Sernin. Neuroscr. 2, 263-275 (1990). 8. Bressler, S. L. Brain Res. 409, 285-293 (1987). 9. Gray, C. M., Kon~g, P., Engel, A. K. & Slnger, W. Nature 338, 334-337 (1989).

l o . Engel. A. K., Konlg, P., Krelter, A. K. & S~nger, W. Scrence 252, 1177-1179 (1991). 11. Engel, A. K., Krelter, A. K.. Kon~g, P. & Slnger, W. Proc. natn. Acad. Scr. U.S.A. 88, 6048-

6052 (1991). 12. Eckhorn. R et al. Brol. Cybern. 80, 121-130 (1988).

13. Freeman, W. J. & van D~jk, B. W. Brain Res. 422, 267-276 (1987). 14. Krelter. A. K. & Slnger, W. Eur. J. Neuroscr. 4, 369-375 (1992). 15. Murthy, V N. & Fetr, E. E. Proc. natn. Acad. Sci. U.S.A. 89, 5670-5674 (1992). 16. Sanes, J. N. & Donoghue, J. P. Proc. natn. Acad. Scr. U.S.A. SO, 447G4474 (1993) 17. Young, M. P., Tanaka, K. & Yamane, S. J Neurophys. 87,1464-1474 (1992). 18. Tovee, M. J. & Rolls, E. T. Neuroreport 3, 369-372 (1992). 19. Glaser. E. M. & Ruchkin. D. S. Princrples of Neurobiological Srgnal Analysrs 168-176

(Academic. New York. 1976). 20. Gray. C. M.. Konlg, P., Engel, A. K. & Slnger, W. Vrs. Neuroscr. 8, 337-347 (1992). 21. Pfurtscheller. G. & Neuper, C Neuroreport 3, 1057-1060 (1992). 22. Gev~ns. A. S. & Bressler, S. L. In Functional Brarn Imaging (eds Pfurtscheller. G. & Lopes

da Silva, F. H.) 99-116 (Hans Huber, Bern, 1988). 23. Gevlns. A. S. et al. Science 235, 580-585 (1987). 24. Llvanov, M. N. Spatial Organrzatron of Cerebral Processes (Wlley. New York, 1977). 25. Sheer, D. E. In Molecular Mechanrsrns m Memory and Learnrng (ed. Ungar, G.) 177-211

(Plenum, New York, 1970). 26. Bressler, S. L & Nakamura. R. In Computation and Neural Systems (eds Eeckman, F. &

Bower, J. M.) 515-522 (Kluwer, Norwell, Massachusetts, 1993).

ACKNOWLEDGEMENTS. We thank S. Kelso and W. Freeman for comments on earller verslons of the manuscript and D. Krleger for help wlth data transfer. Supported by Natlonal lnstltute of Mental Health grants to EEG Systems Laboratory, San Francisco and to the Center for Complex Systems.

NATURE . VOL 366 . 11 NOVEMBER 1993